Wan2GP (Wan for the GPU Poor) by deepbeepmeep is a free, open source app that runs the big open source video models, Wan 2.x, Hunyuan Video, LTX Video, FramePack-style generators and image models like Flux and Qwen Image, on ordinary consumer graphics cards. It does this with aggressive memory management and quantization, so a 12 GB card can generate clips that normally need 24 GB or more. The Gradio interface is a straightforward form: prompt, settings, generate.

Wan2GP is for anyone who wants local AI video without building ComfyUI graphs and without a top-end GPU. It is Windows and Linux only and expects you to be comfortable with a small amount of command line work (or to use Pinokio, which installs it in one click).

Aucune installation requise – exécutez les modèles IA dans le cloud

BitVector Prism est la façon la plus simple de commencer : choisissez un modèle, tapez un prompt et générez dans une application web épurée, sans rien configurer. BitVector est aussi disponible sur Discord et sur le web (SpyGlass).

Essayer BitVector

Ce qu’il vous faut

Windows 10 or 11 or Linux with an NVIDIA GPU: 6 GB VRAM for the smallest models, 12 GB for comfortable Wan 2.x work, 16 to 24 GB for high resolution and longer clips. 32 GB of system RAM is strongly recommended, since models are swapped to RAM. 50 GB or more of disk space for model downloads.

Comment obtenir et installer Wan2GP

Easiest: install Pinokio and pick Wan2GP from its Discover page; it installs everything with one click.
Manual: install Python 3.10 and Git, clone the GitHub repository, create a virtual environment and install PyTorch with CUDA (the README gives the exact commands).
Run pip install -r requirements.txt in the virtual environment. Optional speed-ups like SageAttention are explained in the README.
Start it with python wgp.py. The browser opens on the Gradio page; the model you pick is downloaded automatically from Hugging Face on first use.
Add --profile to choose a memory profile for your card, or pick it in the interface configuration tab afterwards.

Téléchargez uniquement depuis les liens officiels de cette page. Les installateurs trouvés ailleurs peuvent être obsolètes ou modifiés. ModelTrends.ai n’héberge aucun fichier.

Aucune installation requise – exécutez les modèles IA dans le cloud

PirateDiffusion est réservé à Telegram et conçu pour les pros : des milliers de modèles, LoRA et workflows pilotés par commandes de chat, avec génération illimitée à prix fixe.

Essayer PirateDiffusion

Comment utiliser Wan2GP

Choose a model at the top: Wan 2.x text to video, image to video, Hunyuan Video, LTX Video, or an image model such as Flux or Qwen Image.
Write your prompt (and a negative prompt), set the resolution, number of frames and steps. Start small: 480p, 49 frames, 20 to 30 steps.
For image to video, upload a start image; the model animates it following the prompt.
Add LoRAs from the LoRA section (put the files in the loras folder for your model type) and set their multipliers; include the trigger words from the model page.
Press Generate and watch the progress; clips are saved in the outputs folder and shown in the gallery when done. Queue several prompts and let them run.

Conseils et problèmes courants

Choose a lower memory profile if you get out of memory errors; profile 4 or 5 trades speed for lower VRAM.
Distilled or step-accelerated LoRAs (CausVid, LightX2V, FusionX) cut generation to 4 to 8 steps and are the biggest speed win on small cards.
TeaCache and Skip Layer Guidance options speed things up with little quality loss; try them once your settings work.
Keep the project updated with git pull; new models arrive almost weekly.

Modèles utilisables dans Wan2GP

Wan2GP runs the Wan, Hunyuan and LTX video models in this catalog and their LoRAs, plus Flux and Qwen Image for still pictures.

Aucune installation requise – exécutez les modèles IA dans le cloud

Pas de GPU, pas de téléchargement, pas de Python. Ces services exécutent les mêmes modèles open source sur leurs propres serveurs : vous générez depuis un navigateur, Discord ou Telegram en une minute.

Essayer BitVector

Essayer PirateDiffusion

Liens

Ces liens mènent aux sites des éditeurs. ModelTrends.ai n’est pas affilié à eux et n’héberge aucun logiciel.

Autres applis locales

Comment installer ComfyUI

The node-based powerhouse for Stable Diffusion, Flux, Wan video and almost every new open source model. Free, open source, for Windows, Mac and Linux.